Spectra Story
Storycrafting - Northeast Shelf Spectra
I think we need to step back a bit and find the main story again–which is about how size structure of the NES fish community has changed over time, whether that varies by region, and what covariates are associated with those changes.
The Aug draft you developed really focused on comparing regression models to consider drivers based on the fish community size spectrum as viewed from the perspective of the Wigley species set vs. the full species set.
However, I didn’t see figures of those spectra in that draft (maybe there are somewhere else and I’ve lost track). The recent slide deck really focuses on seasonal differences in the size spectra and influence of temperature.
Starting from Square Uno
I’d like for us to start by rebuilding the story of how the size structure has changed over time and regional differences.
As such, I would like for us to look at plots of the size spectrum slope for (1) the all-species length spectrum and (2) the Wigley species biomass spectrum from annual, seasonal and regional perspectives.
I am thinking this would be two panel plots (all-spp length, Wigley-spp biomass) with 5 subpanels each (NES, GOM, GB, SNE, MAB) that are set up to show lines by seasonal and annual time steps, much like the plot on the right of slide 7 in your most recent slide deck…except with NES added and an annual line added. Keeping the significant trend lines in is helpful. This is the first step, so when you have that ready, please share it with the group.
I also think it would be valuable to build out the story of size change further, so am thinking that plots like those of average (or median) length (all species) and weight (wigley species) for NES and the sub-regions, like those on p. 12 of the attached file above (the very first draft from late 2022) would be useful.
Couple things have started to catch my eye the more I’ve thought about volatility in these numbers. There seem to be two situations happening on occassion. The first is more common in GOM+GB, and that is a 5-10cm drop in median length. This sudden drop in size I am imagining is likely due to surges in new recruits.
The other patttern which is more common in MAB but looks like it happens a handful of times in GB is the reverse situation, where median size surges upward in isolated years. For these situations I think we likely could trace these down to exceptionally high catches of larger species like elasmobranchs.
Summary of Trends
Let’s start with these pieces, and we can then discuss doing something with the extremes of the size distribution–changes in small vs. large individuals, where we pre-define size bins as you’ve done in slides 7 and 8, or perhaps using an approach similar to Lora’s approach of defining the upper/lower percentiles of size.
Hope this sounds good as some first steps towards shaping out the results section. And I hope once we see these, we can iterate forward to next steps pretty quickly.
Spiny Dogfish Sensitivity
How much can one species shift the spectra
LArge Fish Index
When I looked at large/small fish indices I looked at two metrics:
- The large fish index (proportion of biomass total represented by fishes in top 5-10% of body sizes)
- The large species index
We will focus on the former for the plots below. Fish size for the below plots will be based on length because these are measured for all individuals.
Large fish will be considered anything over 73cm, representing the 95th percentile.
Values are the fraction of total bodymass for a given season that is coming from the large fishes.
10% 50% 90%
10 21 55 
The fraction of the biomass held in fishes larger than 73cm is lower in the two northern areas, and falling across both seasons.
In SNE its declining in the spring, and in the MAB its increasing.
# A tibble: 37 × 1
comname
<chr>
1 atlantic angel shark
2 atlantic cod
3 atlantic wolffish
4 bluefish
5 cusk
6 goosefish
7 haddock
8 pollock
9 southern stingray
10 spiny dogfish
# ℹ 27 more rowsDistractions
I moved all this stuff below after wasting a bunch of time chasing down loose ends. Enjoy.
Initial Thoughts I’d like to Discuss
Our hypotheses both relate primarily to abundance of non-age0 fishes. It might make sense to move the minimum size up to something larger. This will accomplish two things: 1) lower any noise in median size or spectra slope that results from large recruit cohorts & 2) help ensure that the size range we’re estimating spectra for are fully selected by the gear.
Currently a 1g minimum size is likely lower than the mesh size selectivity.
The following plot shows what the distriution looks like for all data (years, seasons, areas) over the range of sizes we are allowing to contribute to the estimation of seasonal spectra. I’ve highlighted where a minimum body size of 1g sits on this curve to mark where it sits.
Is it even Pareto?
It is quite a pain to determine whether the data follow a pareto distribution in the first place. I’m unsure if the lines in the plots below faithfully follow how they are estimated and how to score goodness of fit for a probability density function.
The MLE method avoids binning and regression. Binning in general can be problematic (e.g. if a data set has no body masses <10 g but the lowest bin is defined as 8–16 g), and the choice of bin widths can affect the estimated slope (Vidondo et al. 1997). Regression-based methods are problematic because the intercept and the slope implicitly determine urn:x-wiley:2041210X:media:mee312641:mee312641-math-0064, which can erroneously be greater than some data values (James, Plank & Edwards 2011). They also assume that the errors in the logarithmic counts for each bin have the same variance, which may not be justified. Although regression can be understood in a likelihood context, this is different to explicitly using a likelihood-based method (Edwards et al. 2012).
After doing some attempts to plot the impacts of moving it around, I am less convinced that its helping things and it may even be harming the situation.
I picked 16g here basically eyeballing things, but my point is mostly that I think there is a need and an advantage to shifting the minimum size. There is a similar lack of tuning for minimum length as well.
We could probably be more scientific and look at the mesh size used, but I think precision there is less important that being closer than we are now.
For maximum size we presently take the maximum individual size in that area-year-season unit. While this also affects the distribution shape I am less concerned with this for now.
What if we shift it?
I’m just going to go ahead and do it and see
Sensitivity to minimum size
Sensitivity to max size
